Don't panic! Just build! https://xitter.cf/t.co/FNuUXBB6HY

Joined December 2013
Lukas Fisch retweeted
The only real moat is giving a shit.
13
78
9
745
37,613
Lukas Fisch retweeted
This really speeds up preprocessing and shows - yet again - that neural networks are eating software. Great work, @codingfisch !! Happy to see this finally published.
1
2
187
Our GPU stack for both NVIDIA and AMD, aside from minimal pieces of signed firmware, is 100% open source and pure Python except for the compiler. It's not using vendor drivers, frameworks, or libraries. That's why it's so easy to make it work on Mac. For compilers, on AMD, we use upstream LLVM, and on NVIDIA, we use the NAK compiler from the MESA project. We plan to replace the compiler with pure tinygrad in a year or two as well. With RANGEIFY merged, our lowering stuff now matches the state of the art, TVM style. We're studying ThunderKittens and TileLang for speed at that level, and should have all this stuff ready in 200 days for the due date of our AMD Llama 405B training contract. Due to tinygrad's small size and pure Python nature, it's the easiest ML library to make progress on, aka fastest slope of improvement. With Megakernel style for scheduling, MODeL_opt style for planning, and E-graph style for symbolic, we should blow past the state of the art in PyTorch and JAX speed. If we do that, NVIDIA's moat is over. It's 1000 lines at most to add a new accelerator to tinygrad. And I don't mean to add a new accelerator with help from a kernel driver, compiler, and libraries. Just 1000 lines of software for the *whole* accelerator speaking right on the PCIe BARs, like what tinygrad is doing with the NVIDIA and AMD GPUs now.
24
55
11
819
61,051
Lukas Fisch retweeted
The curse of dimensionality. The blessing of compositionality.
6
19
3
274
24,252
Lukas Fisch retweeted
François Chollet
@fchollet7 Sep 2025
When you store your knowledge and skills as parametric curves (as all deep learning models do), the only way you can generalize is via interpolation on the curve. The problem is that interpolated points *correlate* with the truth but have no *causal* link to the truth. Hence hallucinations. The fix is to start leveraging causal symbolic graphs as your representation substrate (e.g. computer programs of the kind we write as software engineers). The human-written software stack, with its extremely high degree of reliability despite its massive complexity, is proof of existence of exact truthiness propagation.
62
121
23
1,216
114,489
Couldn't agree more with Linus here. Making your code running as fast as possible (or just faster than someone's else code) is the best feeling you can get from programming The issue that it is extremely hard to find the project that will pay you for instruction level efficiency
69
340
46
4,484
417,149
Lukas Fisch retweeted
François Chollet
@fchollet13 Jul 2025
Our thesis at Ndea: simple theories should be discoverable using little data and little compute. And everything humans have ever invented is a fairly simple composition of fairly simple theories.
AI can't figure out the inverse square law looking at 10M solar systems, Newton figured it out looking at 1
41
61
9
899
136,345
Lukas Fisch retweeted
NCCL sending the loss value from the last pipeline parallel stage back to rank 0 so the user can print it
6
20
310
19,893
I’m speaking at 2pm on Wednesday at the Upper Bound AI conference. I am in the “business of AI” track, but my talk will be about reinforcement learning research. Video should be available after the conference.
40
56
10
1,338
124,290
40
161
28
1,466
90,188
Great software isn't invented, it's discovered.
22
22
5
462
34,681
Replying to @wwu_mmll
@wwu_mmll is looking for a PostDoc (3 years with possible extension). Work will be at the intersection of ML, dynamical systems, and neuroscience. Great team, beautiful city. DM me if you did your PhD on ML/dynamica systems and are interested in working in neuroscience.
1
5
2
3
1,504
Lukas Fisch retweeted
Zooming into an iPhone CPU die (5nm) ⁽ᵛᶦᵈᵉᵒ ᵉᵈᶦᵗᵉᵈ ᶠᵒʳ ᵉᵈᵘᶜᵃᵗᶦᵒⁿᵃˡ ᵖᵘʳᵖᵒˢᵉˢ⁾
261
2,372
271
14,014
1,419,530
Bernhard Riemann died in 1866 at the age of 39. Here is a list of things named after him. Riemann bilinear relations Riemann conditions Riemann form Riemann function Riemann–Hurwitz formula Riemann matrix Riemann operator Riemann singularity theorem Riemann surface Compact Riemann surface The tangential Cauchy–Riemann complex Zariski–Riemann space Cauchy–Riemann equations Riemann integral Generalized Riemann integral Riemann multiple integral Riemann invariant Riemann mapping theorem Measurable Riemann mapping theorem Riemann problem Riemann solver Riemann sphere Riemann–Hilbert correspondence Riemann–Hilbert problem Riemann–Lebesgue lemma Riemann–Liouville integral Riemann–Roch theorem Arithmetic Riemann–Roch theorem Riemann–Roch theorem for smooth manifolds Grothendieck–Hirzebruch–Riemann–Roch theorem Hirzebruch–Riemann–Roch theorem Riemann–Stieltjes integral Riemann series theorem Riemann sum Riemann–von Mangoldt formula Riemann hypothesis Generalized Riemann hypothesis Grand Riemann hypothesis Riemann hypothesis for curves over finite fields Riemann theta function Riemann Xi function Riemann zeta function Riemann–Siegel formula Riemann–Siegel theta function Free Riemann gas Riemann invariant Riemann–Cartan geometry Riemann–Silberstein vector Riemann-Lebovitz formulation Riemann curvature tensor Riemann tensor Riemannian graph Riemannian group Riemannian holonomy Riemannian manifold also called Riemannian space Riemannian metric tensor Riemannian Penrose inequality Riemannian polyhedron Riemannian singular value decomposition Riemannian submanifold Riemannian submersion Riemannian volume form Riemannian wavefield extrapolation Sub-Riemannian manifold Riemannian symmetric space Riemann's differential equation Riemann's existence theorem Riemann's explicit formula Riemann's minimal surface Riemann's theorem on removable singularities
27
327
58
1,672
191,076
📢 Exciting News! Our recently published paper explores the link between childhood maltreatment, social support, and functional activity during negative emotion processing. Shared first authorship with @AnneRinck_
1
4
2
7
1,889
🧠🔬JOB Alert - PhD-position 📢 Join our fantastic team @TraPLabMS and let's find out how replicable neuroimaging research in depression really is! Why do some brain findings replicate while others don't? Excited to receive your application by April 10 via tinyurl.com/2dl5qrjp
1
12
6
11
3,856
Those who would give up essential Simplicity to purchase a little temporary Speed deserve neither Simplicity nor Speed.
7
12
3
213
20,473
📢Publication Alert! Check out my new article on long-term effects of electroconvulsive therapy on brain structure in depression published in Psychological Medicine!⚡️🧠 doi.org/10.1017/S00332917230… 🧵
1
6
1
14
2,414
Had an awesome #TraPLab_Münster lab conference in the Münster University castle
3
7
16